2022 International Conference on Advancement in Electrical and Electronic Engineering 24-26 February 2022, Gazipur, Bangladesh. P-205 Identification of Vector and Non-vector Mosquito Species Using Deep Convolutional Neural Networks with Ensemble Model 1 st Md.Abedur Rahman Shamim Dept.of CSE University of Barishal Barishal,Bangladesh Email: 047shamim@gmail.com 2 nd A.B.M Anas Dept.of CSE University of Barishal Barishal,Bangladesh Email: anas14cse@gmail.com 3 rd Md.Erfan Dept.of CSE University of Barishal Barishal,Bangladesh Email: irfan.bucse@gmail.com Abstract—Human life has always been suffering from insects, particularly mosquitoes since its early beginnings. This annoying insect acts as a vector that transmits pathogens by feeding on our blood, spreading critical diseases like Zika Virus, Malaria, dengue fever, Chikungunya, etc. It’s important to stop these dipterous insects from harming humans and need a method to identify the vector species. For many years, image-based automated identification of vector mosquitoes has been studied for applications such as early identification of mosquito-borne diseases. Here Deep Convolutional Neural Networks (DCNNs) are modern-day techniques for extracting visible functions and classifying objects and, there exists an excellent application for the classification of images. In this study, we analyzed the functionality of deep learning models in classifying mosquito species having excessive inter-species similarity and intra-species variations. We constructed a data set with approximately 3600 images of eight mosquito species with diverse postures and deformation conditions. Our result demonstrated that more than 98% classification accuracy has been achieved by using our proposed ensemble method on this data. We also showed the comparison of various DCNNs models such as VGG-16, InceptionV3, and MobileNetV2. The overall results show that InceptionV3 is the best model with 99.38% of training accuracy and 97.02% of testing accuracy. Index Terms—Convolutional Neural Networks, Deep Learning, Transfer Learning, VGG16, InceptionV3, MobilenetV2 I. I NTRODUCTION Mosquito-borne diseases such as malaria, dengue, West Nile, and Zika viruses represent serious public health problems with significant human and economic costs. For example, malaria alone kills more than one million people worldwide each year, most of them are children [1]. In Southeast Asian countries such as Bangladesh, Malaysia, Philippines, and Vietnam, 2019 was the year with the highest dengue fever [2]. There is usually no vaccine or treatment for these diseases and prevention is based on mosquito monitoring and control. This transition requires accurate knowledge of real-time geographic presence. There are about 4500 mosquito species (common to 34 genera) where Aedes (Ae.), Anopheles (An.), and Culex (Cu.) are spreading diseases [6]. There are several species within this genus. Malaria is largely spread by Anopheles gambiae in Africa and Anopheles stephensi in India. Aedes aegypti is the main carrier of dengue, yellow fever, chikungunya, and Zika viruses. West Nile and other encephalitis viruses are transmitted by Culex nigripalpus. Bangladesh is a densely populated country and most of the people are living in unhealthy places. Every year, hundreds of People died from mosquito-borne diseases and thousands got sick. Dengue fever, malaria, and chikungunya are the most common diseases. But in recent years, the problem has become serious. In 2019, at least 18 people died from dengue fever and 16,223 were infected [3]. In 2020, amidst the pandemic, a total of 1026 confirmed cases were reported by the Directorate General of Health in Bangladesh [4]. The purpose of this study is to create an ensemble model ca- pable of recognizing several mosquito species such as Aedes, Anopheles, and Culex from a given input image. The aim would be achieved by training a model on our dataset using Transfer Learning techniques. The objective of the research is to improve the accuracy and compare it to other systems. As a result, in this research, a study of multiple existing Transfer Learning Models such as VGG16, InceptionV3, and MobileNetV2 are conducted and evaluated in terms of accu- racy, precision, Recall, F1 Score, and training time, in order to build an effective and efficient automated classifying system. II. RELATED WORK Mosquito classification has attracted the interest of many researchers. Recent studies proposed several classification techniques, but efficiency and computing complexity remain a tradeoff in the majority of research. In the paper by Krizhevsky et al. [5], exhibited a record-breaking performance in the ImageNetLSVRC-2012 competition and achieved top-1 and top-5 error rates of 39.7% and 18.9%. The researchers trained a Convolutional Neural Network Model (AlexNet) using the ImageNet dataset, which comprises over 1.2 million photos divided into 1000 categories. Pre-processing included scaling photos to 256x256 resolution, followed by further data aug- mentation to produce 224x224 images. These pictures were 978-1-6654-6944-9/22/$31.00 ©2022 IEEE